![Sample our Economics, Finance,Business & Industry journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5931/TOC-Subject-Sample-2021-Economics-Finance-Business-Industry.jpg"})
Abstract
The literature on exchange rate regimes has paid little attention to the effects of exchange rate policies on real exchange rate misalignments. This paper contributes to filling that gap by exploring such relation empirically. Because the underlying model is probably not linear and the treated individuals differ from non-treated individuals, we rely on Matching models rather than on standard regressions. Our main finding is that pegs are associated with more overvaluation. The results are robust to different exchange rate regime classifications, misalignment indexes, and matching estimators. The evidence presented suggests that policy-makers concerned with overvaluation should avoid sticking with rigid arrangements for too long.
Acknowledgements
We would like to thank Martín Rapetti, Roberto Frenkel, Mario Damill, Gabriel Palazzo, Diego Friedheim, Constanza Albuín, Simon Sturm, and an anonymous referee for comments and suggestions. They should not be blamed for any remaining errors.
Notes
1. In this paper we use the Latin-American convention, so an increase (decrease) in the exchange rate implies a devaluation (revaluation).
2. We will refer to ‘positive misalignment’ (when the RER is above the ‘equilibrium RER’) as ‘undervaluation’ and to ‘negative misalignment’ (when the RER is below the ‘equilibrium RER’) as ‘overvaluation’.
3. We are indebted to an anonymous referee for pointing out such reference.
4. We will refer broadly to ‘matching’ to include models that use the propensity score and models that do not, while we will refer to PSM when only models that use the propensity score are included.
5. See e.g. Rodrik (Citation2008), Eichengreen (Citation2008), and Korinek and Serven (Citation2010).
6. Standard regressions do not require any assumption regarding the distribution of the variables, but normality is imposed for inference.
7. In fact OLS is a particular form of matching. To put it differently, PSM is like OLS with a different set of weights: while PSM puts most of the weight on those individuals who are more likely to be treated, a standard regression puts most of the weight where the conditional variance of the treatment status is the largest; roughly speaking, where the probability of being treat is equal to the probability of not being treated (or just one half).
8. See Angrist and Pischke (Citation2009) for a formal presentation of the hospitalization example.
9. Notice that in the ERR literature the propensity to adopt a particular regime is often estimated, and the effects of the ERRs on the economy are usually analyzed. But the estimations of the propensity score and the impacts of different regimes are part of different papers. PSM can easily combine both approaches.
10. The indexes can be classified according to: (a) the nature of the data (panel vs. time-series); (b) single equations vs. multiple equations approaches; and (c) the underlying model. For an excellent survey, see the book by Hinkel and Montiel (Citation1999).
11. E.g. Berg and Miao (Citation2010), McDonald and Vieira (Citation2010). They study the relation between misalignment and growth.
12. We use five years average to diminish the concern with endogeneity issues, both for the trade concentration index and the trade openness measure. Notice that a higher level for the Theil index means lower diversification.
13. The main results using the RR classification holds if we include Dollar instead of Foreign_liab.
14. These results were obtained using the Stata command PSMATCH2 developed by Leuven and Sianesi (Citation2003).
15. Once we match individuals with similar propensities to adopt each ERR, the pseudo R2 should vanish because the covariates will no longer explain regime choice (if the process of matching is accurate).
16. The mean standardized bias is defined as , where the X represent the mean of the control or non-treated units (C) and the treated units (T), and the S2 represents the the variances (also of the non-treated and treated units).
17. This index is a naive but reasonable and simple attempt to track deviations from Purchasing Power Parity. That is why we name it ‘PPP’.
18. Interestingly, the properties of the estimations using the IMF classification are superior our baseline results using RR. For instance, the covariate imbalance using the Mahalanobis distance looks pretty good after matching (results available upon request).